Powerboost 1000c does not recharge

[Johannes_7]

For some reason, powerboost 1000c will not recharge the batteries (V_Bat, 4x18650 with 12Ah). The graphic below shows V_Bat (green line) decreasing over a week from 4.2V to 3.0V due to some consumers, that are directly connected to the batteries. The blue line (V_EXT) shows, that external power supply is always present.
On the 27th of May,

• I turned the device off and on
  powerboost checks the battery level is low
  and starts charging for the next 24h.
  Since the 28th of May V_BAT is going down again.

I suspected the security Timer of the MCP73871, but *TE is connected to high, so it should be disabled.

BatDrain.png (45.12 KiB) Viewed 555 times

As I understood the datasheet, V_BAT should be evaluated regularly, but apparently it is not. Is there a pin I can set from the µC to trigger a new evaluation of V_Bat? Or is my last chance to install a time-switch on the powersupply to interrupt V_EXT once a day and thus make the MCP73871 check V_BAT?

[adafruit_support_mike]

Check your load current.

The MCP73871 routes power to the boost converter first, and charges the LiPo with any supply current not being used by the load.

The maximum current through the MCP73871 is 1800mA, which scales down to about 1.1A @ 5V for the load.

[Johannes_7]

The load current @ 5V (VDD) is only 60mA.
The load current @ V_BAT is 55mA for 5 minutes and 150mA for 20 seconds.
I am surprised, that the MCP73871 even reaches the state "charge-complete", presumably through the programmed termination current, which is 24.75mA to 41.25mA according to viewtopic.php?f=19&t=125692&p=627632#p627876.
A second test-setup, which I set to 5mA @ V_BAT for 30 seconds and 150mA for 20 seconds never (that is 16h) reaches the state "charge-complete", i.e. the yellow led is always on. However, I only had one full 18650 in the battery tray.
Just updated the test to 4x 18650; however I don't expect any change here.

[adafruit_support_mike]

The PowerBoost 1000C is designed to consider 'full charge' the point where the current flowing into the battery under constant-voltage conditions falls below 10mA.

That can take a very long time with LiPos that have high storage capacity. The constant-voltage phase works more or less like water diffusing into a sponge: if you treat the polymer as a stack of thin sheets, each layer absorbs charge from the one below it at a rate proportional to the difference between their charges. Charge moves slowly when the difference between two sheets is small, but the absolute amount of charge that moves can be fairly large if the sheets are large.

You can change the full-charge current by replacing the 100k resistor connected to pin PROG3:

2465-06.jpg (84.11 KiB) Viewed 511 times

The cutoff current is 1000 times the current through the resistor at 1V.

It's common to define full charge as the point where the current flowing into the LiPo is 1% of the 1C discharge rate, which would be 120mA for a 12Ah battery. That would correspond to an 8.2k resistor.

[Johannes_7]

Thank's for clarifying the programmability of the termination current. My second test-setup with 12Ah and 5mA constant drain with 150mA peaks every 5min through a consumer on V_BAT didn't reach "charge-complete", during the last 24h. There is absolutely no voltage drop on V_BAT and the yellow led is still on. But this is perfect and doesn't cause any problems.
The issue is that in my first setup, once the MCP73871 reaches "charge complete", as indicated in download/file.php?id=84097, it doesn't check V_BAT anymore, leaves it draining out over a week, never checks V_BAT again until I cut the power supply and turn it back on.
Is it illegal to power consumers directly from V_BAT, so that the batteries are "drained behind the MCP73871's back"? Should I only power through the powerboost's 5V_OUT (5.2V, "VDD") and use a separate buck converter to create 3.3V from 5V_OUT?
The problem is, that this application is already installed at a remote place and access requires the client.

Another solution would be to install a timer-switch and cut V_EXT every day. We all know this kind of solution; but if possible I would like to understand what's going on.